The author questions some of the fundamental concepts of the classical theory of measurement and theory of errors. Rather than treating measures as real numbers, a measure of a parameter should be a full value-span, i.e., a (fuzzy) subset of numbers tied to a unit of measure, meant to represent as a whole the parameter. Its width, which embodies the uncertainty of the measure, is assigned so as to ensure that the various measures of the same parameter are congruent, i.e., their value-spans overlap. In sum, the classical mapping of empirical entities into the set of real numbers should be replaced by a more general relational structure between empirical entities, measures, and reading-values.